AI agents that work with your tools and data
An AI audit helps us choose a workflow worth improving and a KPI for checking the results. We then build an AI pilot with human review and a clear fallback. The pilot deadline is written into the proposal and guaranteed.
The problem we solve
Many teams already use AI tools without turning them into systems that improve a specific process. A generic chatbot does not fix a bottleneck, and automating the wrong task only creates more activity without a clear business result.
Who this is for
- Owners and operations leaders who know where a process gets stuck but do not want to build AI on guesswork
- Sales, support and administrative teams handling repetitive enquiries, documents or follow-up
- Businesses that want a small pilot with a clear KPI before investing more in AI
- Teams that need human review, records and a fallback before AI interacts with customers
What we build
How the workflow works in practice
Illustrative examples with fictional data. They show the process and human review, not the results of a client project.
From an enquiry to a draft reply
- Input
Fictional enquiry: We need a website with an enquiry form. Can you send us a price and delivery date? Approved information: The proposal depends on the confirmed scope and required materials. The price and delivery date for this enquiry have not been agreed.
- Output
Thank you for your enquiry. To prepare a proposal, we need the number and types of pages, the form features you require, and confirmation of whether your copy and photographs are ready. Once the scope is confirmed, we can propose a price and delivery date.
- Process
- Identify the requested service and the information still needed for a proposal.
- Prepare a draft using the approved information.
- Hand the draft to the person who reviews the reply and decides whether to send it.
- Human review
- Check that the draft matches the actual enquiry and approved offer.
- Confirm the facts, tone and questions for the customer. A person approves and sends the reply.
- If information is missing or sources conflict, return the draft for revision.
- What we measure
- Total time spent preparing and reviewing the reply.
- Number of factual corrections and situations that needed further clarification.
Sample AI audit findings
An illustrative review of a fictional enquiry-reply process. This is not an assessment of your company or a measured result.
- Accountability
- The person responsible for enquiries reviews and approves the reply. Before the pilot, that person and a backup must be named.
- Data readiness
- Approved service descriptions are available, but a price or delivery date for a particular enquiry may be missing. Sources, access and the process for filling gaps must be confirmed.
- Process
- AI prepares a draft, and a person decides whether to send it. Review steps, exceptions and the manual fallback if processing fails must be documented.
- The baseline has not been measured
- Time to an approved draft and the number of corrections are not yet known. They must first be recorded on an agreed sample, including the time spent on human review.
- Expected benefit
- The hypothesis is less rewriting and a clearer list of unknowns. The pilot must test that benefit alongside reply quality and the total cost of the work.
- Recommendation and next step
- A limited drafting pilot makes sense once sources and responsibilities are agreed. Without those conditions, the data and process need to be put in order first.
How we choose the priority
We consider business impact and feasibility. A recurring task may justify a pilot if sources are accessible, someone is accountable and there is a clear way to check the result. A large expected benefit alone is not enough to start. We agree the actual priority after reviewing the process and measuring the baseline.
How we evaluate an AI pilot
Illustrative criteria for the same fictional process. They show what we would test, not the results of a completed evaluation. We agree specific cases and pass thresholds before the pilot.
- The facts are available
- The draft answers the enquiry using approved sources. A person can check every claim against its source and confirms the facts and tone.
- The price or delivery date is missing
- The draft identifies what is missing and prepares a question to clarify it. It does not invent a price, delivery date or promise to the customer.
- The sources conflict
- The system flags conflicting data and asks for a human decision. The draft does not proceed until the responsible person confirms which information applies.
- The system is unavailable
- Failed or excessively slow processing returns the enquiry to the responsible person for the agreed manual process. The enquiry remains recorded and is not marked as resolved.
- There is no human approval
- The reply is not sent to the customer. The test checks the sending action as well as the draft text.
Who decides what happens next
Before the pilot, we name the person who confirms the business benefit and the person responsible for technical checks and corrections. Together, they review test results and measurements and record a decision.
- Expand
- The agreed quality, safety and benefit criteria are met, and the team uses the process in its work.
- Improve
- There is a specific gap, a person responsible for correcting it and a test that will confirm the improvement before expansion.
- Stop
- There is no clear benefit, access rules are not met, or a safer process handles the task better.
Why DTC
We first assess the process, tools and data, then choose AI, automation or a simpler solution.
We agree a measurable pilot result: response time, missed enquiries, manual steps, errors or processing speed.
We build approvals, logs, fallback and reporting for the owner or team leader into the system.
After the audit, we write the pilot build deadline into the proposal. If we miss it through our own fault, you do not pay for the pilot build.
After the pilot, you receive an honest conclusion: expand, improve or pause. If straightforward automation is better than AI, we will say so.
AI, automation, websites and custom software come from the same team, so responsibility is not passed around.
How the offer works
AI audit
We assess the workflow, tools and data. You receive a pilot recommendation, an agreed KPI and a list of required integrations.
AI pilot
We build a bounded AI workflow with human review, logs and fallback. Results are compared with the baseline.
AI operations
We monitor the system and KPI, maintain prompts and integrations, and agree improvements. The owner or team leader receives a view of the results.
Processes worth reviewing
How we check the result
We record how the process works today: steps, duration, errors, manual work and the process owner.
We agree one meaningful number to measure the pilot, rather than relying on general impressions of AI.
We run a bounded workflow on a real process, with clear review and fallback.
We compare outputs, manual corrections, time and situations where AI must not decide on its own.
You receive a results summary and a recommendation: expand, improve, pause or use simpler automation.
Boundaries before AI
- !We do not promise revenue, rankings or productivity without measurement.
- !We do not release customer-facing automation without human review and fallback.
- !We do not work with sensitive data before agreeing access, retention and security rules.
- !We do not build AI when a simpler workflow or deterministic automation would solve the bottleneck better.
Let’s start with one workflow.
AI audit · guaranteed pilot deadline · reply within 24h
Frequently asked questions
A chatbot is just one possible format. We start with the workflow and KPI, then decide whether it needs an agent, a RAG assistant, automation, data validation or a custom integration.
We use OpenAI, Anthropic, open-source models, n8n, Make, Zapier or custom code according to the problem. Tools are chosen after the audit, not before.
Not in the v1 pilot. Critical outputs go through human review, and the workflow must have logs, fallback and clear stop rules.
We begin with an AI audit and then define the pilot scope. Price depends on the data, integrations, risk and how much human review must remain in the process.
After the audit, we write the pilot build deadline into the proposal. If we miss it because of something on our side, you do not pay for the pilot build. If we are waiting for your access, data or approvals, the deadline moves by the number of days we waited — and we tell you straight away.